Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Data driven modeling for system-level condition monitoring on wind power plants

: Eickmeyer, Jens; Li, Peng; Givehchi, Omid; Pethig, Florian; Niggemann, Oliver

Fulltext urn:nbn:de:0011-n-3642998 (270 KByte PDF)
MD5 Fingerprint: 76f4cc28e605ead4a136c2ddbb627148
Created on: 17.11.2015

Pencolé, Yannick (Ed.); Travé-Massuyès, Louise (Ed.); Dague, Philippe (Ed.):
26th International Workshop on Principles of Diagnosis, DX 2015. Proceedings. Online resource : August 31-September 3, 2015, Paris, France; Co-located with 9th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes, Safeprocess 2015
Paris, 2015 (CEUR Workshop Proceedings 1507)
URN: urn:nbn:de:0074-1507-1
International Workshop on Principles of Diagnosis (DX) <26, 2015, Paris>
Symposium on Fault Detection, Supervision and Safety for Technical Processes (SAFEPROCESS) <9, 2015, Paris>
Conference Paper, Electronic Publication
Fraunhofer IOSB ()

The wind energy sector grew continuously in the last 17 years, which illustrates the potential of wind energy as an alternative to fossil fuel. In parallel to physical architecture evolution, the scheduling of maintenance optimizes the yield of wind power plants. This paper presents an innovative approach to condition monitoring of wind power plants, that provides a system-level anomaly detection for preventive maintenance. At first a data-driven modeling algorithm is presented which utilizes generic machine learning methods. This approach allows to automatically model a system in order to monitor the behaviors of a wind power plant. Additionally, this automatically learned model is used as a basis for the second algorithm presented in this work, which detects anomalous system behavior and can alarm its operator. Both presented algorithms are used in an overall solution that neither rely on specialized wind power plant architectures nor requires specific types of sensors. To evaluate the developed algorithms, two well-known clustering methods are used as a reference.